Last week, on Thursday 27 April, the European Parliament agreed on a provisional AI Act. The legislation is subject to amendments as it makes its way through the trilogue process, but in its current iteration, it lays out a provisional framework for the classification of AI tools by risk and obliges generative AI tools to disclose copyrighted material used for “training”.
Penalties for non-compliance with the AI Act are substantial, with fines capped at the higher of 30 million euros or 6% of global income.
According to a Reuters report, disclosure was preferred to an outright ban on using copyrighted material for algorithm training advocated for by some committee members. The fact that the most up to date draft of the current legislation is not yet available limits our ability to comment constructively on current provisions, but whether this flexible approach will work better than a bright line rule is uncertain.
Whilst the issue of credit may be addressed by a disclosure requirement, the commercial risk to publishers presented by user products like ChatGPT diverting traffic, and thus advertising revenue, from content owners will plainly be unaffected.
Notwithstanding this, it is certainly positive to see the EU demonstrate flexibility on this issue. The common position recently adopted by the Council in December 2022, indeed, made no mention of the issue of copyright, focusing instead solely on the categorisation and regulation of AI tools determined by their threat level.
Expanding the scope of the legislation, signifies the Commission’s willingness to adapt its approach in the face of a rapidly evolving fact base. MOW will be following developments with interest and will make sure to convey our views to the EU lawmakers.